[Insight-users] MutualInformationEuler2DRegistration

[Ying Du]

Hi Luis,

Thank you so much for the help. I tried the program, it works beautifully on
those jpeg images.

One question is that when I tried it on the same images but with the "TIF"
format, the result is quite different and not good. Does TIF has some
problem here? The real medical images I will be working on are all in "TIF"
format now, do I need to convert them to jpg?

I am using InsightToolkit-1.6.0, and I did not find
"ImageRegistration13.cxx" under Examples/Registration, are the new examples
like this one appear in the CVS version?

Thanks again!

Ying


----- Original Message ----- 
From: "Luis Ibanez" <luis.ibanez at kitware.com>
To: "Ying Du" <duying at hotmail.com>
Cc: <insight-users at itk.org>
Sent: Thursday, July 01, 2004 3:22 PM
Subject: Re: [Insight-users] MutualInformationEuler2DRegistration


>
> Hi Ying,
>
> The best remedy agains "parameter sensitivity" is "user-training"  :-)
>
> What we use to call 'sensitivity' simply reflects the uncertainty
> on the ranges and values that should be used for configuring the
> registration. The solution is simply for you to follow the examples
> in the SoftwareGuide, that introduce you progressively to the
> trade-offs involved in Image Registration. This training will help
> you understand how to fine tune the parameters of registration
> algorithms in order to get reasonable results.
>
>
> Please look at the CenteredRidig2D example in the Software Guide
>
>             http://www.itk.org/ItkSoftwareGuide.pdf
>
> Section 8.5, pdf-page 263.  The source code associated with this
> example is in
>
>               Insight/Examples/Registration/
>                             ImageRegistration5.cxx
>
> You will find there a discussion on the importance of selecting
> the center of rotation for the image.
>
> Note that your images have the interesting characteristic of
> having a center of rotation that is far from the corner of the
> image *and* far from the center for the image.
>
> It is very important that you initialize the center of rotation
> of the transform. That will reduce a lot the sensibility to
> the initial parameters of the registration.
>
> Attached you will find a source code file that registers your
> images in 200 iterations (about 7 seconds in a Pentium 4 at 2.4Ghz)
>
> Note that it is using MattesMutualInformation and the Centered
> transform initializer class.
>
> The same file has been commited as a new example in
>
>
>    Insight/Examples/Registration/
>                 ImageRegistration13.cxx
>
>
>
>
> Please let us know if you have further questions.
>
>
>   Regards,
>
>
>      Luis